Reproduce “Training People to Reward Robots” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Endong Sun, Yuqing Zhu, Matthew Howard
King's College London researchers developed a scaffolded machine teaching framework to train novice users to provide effective reward demonstrations for robots. The approach reduced demonstration errors by 83% and improved robot learning outcomes by 98%, with skills transferring to new tasks.
Key insights
- Machine Teaching can effectively upskill novice human users, enabling them to provide higher-quality reward demonstrations for robot learning.
- Improvements in human teaching quality directly lead to substantial gains in robot learning efficiency and performance.
- Teaching skills acquired through the MT framework generalize across different tasks, demonstrating transferable knowledge.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.